Fig. 1: TranSiGen’s architecture and application. | Nature Communications

Fig. 1: TranSiGen’s architecture and application.

From: Deep representation learning of chemical-induced transcriptional profile for phenotype-based drug discovery

Fig. 1

a The data processing flow for TranSiGen. b The architecture and inference process of TranSiGen. c The applications of TranSiGen-derived representation. \({X}_{1}\) represents the control profile treated with DMSO, \({X}_{2}\) represents the transcriptional profile treated with the compound, \({\hat{X}}_{1}\) represents the reconstructed control profile, \({\hat{X}}_{2}\) represents the reconstructed transcriptional profile, \({X}_{2}^{\prime}\) represents the predicted transcriptional profile, \(\varDelta X^{\prime}\) represents the predicted differential expression profile, \({Z}_{1}\) represents the latent representation of \({X}_{1}\), \({Z}_{2}\) represents the latent representation of \({X}_{2}\), \({Z}_{{mol}}\) represents the hidden representation of the compound, \({Z}_{2}F{Z}_{1}\) represents the latent representation from \({X}_{1}\) and perturbation representation, \({{encoder}}_{x1}\) represents the encoder for \({X}_{1}\), \({{decoder}}_{x1}\) represents the decoder for \({X}_{1}\), \({{encoder}}_{x2}\) represents the encoder for \({X}_{2}\), and \({{decoder}}_{x2}\) represents the decoder for \({X}_{2}\).

Back to article page